arXiv subjects

Zehua Cao

Publications and source records attributed to Zehua Cao.

6 recordsLinked to original sources

arXiv2026

Faithful-Patchscopes: Understanding and Mitigating Model Bias in Hidden Representations Explanation of Large Language Models

Large Language Models (LLMs) have demonstrated strong capabilities for hidden representation interpretation through Patchscopes, a framework that uses LLMs themselves to generate human-readable explanations by decoding from internal hidden representations. However, our work shows that LLMs tend to rely on inherent linguistic patterns, which can override contextual information encoded in the hidden representations during decoding. For example, even when a hidden representation encodes the contextual attribute "purple" for "broccoli", LLMs still generate "green" in their explanations, reflecting a strong prior association. This behavior reveals a systematic unfaithfulness in Patchscopes. To systematically study this issue, we first designed a dataset to evaluate the faithfulness of Patchscopes under biased cases, and our results show that there is an 18.84\% faithfulness decrease on average. We then propose Bias Alignment through Logit Recalibration (BALOR), which treats the output logits from an unpatched prompt as capturing model bias and contrasts them with logits obtained under patched contextual information. By recalibrating the logit distribution through this contrast, BALOR suppresses model bias and amplifies contextual information during generation. Experiments across multiple LLMs demonstrate that BALOR consistently outperforms existing baselines, achieving up to 33\% relative performance improvement.

cs.CL↗

arXiv2025

The $Zα^2$ correction to superallowed beta decays in effective field theory and implications for $|V_{ud}|$

Superallowed ($0^+\rightarrow0^+$) beta decays currently provide the most precise extraction of quark mixing in the Standard Model. Their interpretation as a measurement of $|V_{ud}|$ relies on a reliable first-principles computation of QED radiative corrections expressed as a series in $Zα$ and $α$. In this work, we provide the first model-independent result for two-loop, $O(Zα^2)$, long-distance radiative corrections where the nuclei are treated as heavy point-like particles. We use renormalization group analysis to obtain new results at $O(Zα^3)$ for the coefficient of double-logarithms in the ratio of the maximal beta energy to the inverse nuclear size, $\Em/R^{-1}$. We use the Kinoshita-Lee-Nauenberg theorem to obtain new results at $O(Z^2α^3)$ for the coefficient of logarithms in the ratio of maximal beta energy to the electron mass, $\log(2\Em/\me)$. We identify a structure-dependent, and therefore short-distance, contribution to the traditional $Zα^2$ correction that should be revisited.. We provide the first comprehensive update to the long-distance corrections in almost forty years and comment on the impact of our findings for extractions of $|V_{ud}|$. We find that shifts in the long-distance corrections are $2.5\times$ larger than past estimates of their uncertainty, $1.5\times$ larger than the statistical uncertainty from the combined fit of superallowed decays, and about $1/2$ the size of estimated systematic error, which stems dominantly from nuclear structure effects.

hep-ph↗

arXiv2025

The Fermi function and the neutron's lifetime

The traditional Fermi function ansatz for nuclear beta decay describes enhanced perturbative effects in the limit of large nuclear charge $Z$ and/or small electron velocity $β$. We define and compute the quantum field theory object that replaces this ansatz for neutron beta decay, where neither of these limits hold. We present a new factorization formula that applies in the limit of small electron mass, analyze the components of this formula through two loop order, and resum perturbative corrections that are enhanced by large logarithms. We apply our results to the neutron lifetime, supplying the first two-loop input to the long-distance corrections. Our result can be summarized as \begin{equation*} τ_n \times |V_{ud}|^2\big[1+3λ^2\big]\big[1+Δ_R\big] = \frac{5263.284(17)\,{\rm s}} {1 + 27.04(7)\times 10^{-3} }~, \end{equation*} with $|V_{ud}|$ the up-down quark mixing parameter, $τ_n$ the neutron's lifetime, $λ$ the ratio of axial to vector charge, and $Δ_R$ the short-distance matching correction. We find a shift in the long-distance radiative corrections compared to previous work, and discuss implications for extractions of $|V_{ud}|$ and tests of the Standard Model.

hep-ph↗

arXiv2025

Factorization and resummation of QED radiative corrections for neutron beta decay

Details of the two-loop analysis of long-distance QED radiative corrections to neutron beta decay are presented. Explicit expressions are given for hard, jet, and soft functions appearing in the factorization formula that describes the small mass/large energy limit. Power corrections, cancellation of singularities in the small mass expansion, renormalization scheme dependence, and bound state effects are discussed. The results impact the determination of $|V_{ud}|$ from the measured neutron lifetime.

hep-ph↗

arXiv2021

Qualification tests of 997 8-inch photomultiplier tubes for the water Cherenkov detector array of the LHAASO experiment

The Large High-Altitude Air Shower Observatory (LHAASO) is being built at Haizi Mountain, Sichuan province of China at an altitude of 4410 meters. One of its main goals is to survey the northern sky for very-high-energy gamma ray sources via its ground-based water Cherenkov detector array (WCDA). 900 8-inch photomultiplier tubes (PMTs) CR365-02-1 from Beijing Hamamatsu Photon Techniques INC. (BHP) are installed in the WCDA, collecting Cherenkov photons produced by air shower particles crossing water. The design of the PMT base with a high dynamic range for CR365-02-1, the PMT batch test system, and the test results of 997 PMTs are presented in this paper.

physics.ins-det↗

arXiv2018

Photoproduction of J/$ψ$ in non-single-diffractive p+p collisions

Recently, significant enhancements of J/$ψ$ production at very low transverse momenta were observed by the ALICE and STAR collaboration in peripheral hadronic A+A collisions. The anomaly excesses point to evidence of coherent photon-nucleus interactions in violent hadronic heavy-ion collisions, which were conventionally studied only in ultra-peripheral collisions. Assuming that the coherent photoproduction is the underlying mechanism which is responsible for the excess observed in peripheral A+A collisions, its contribution in p+p collisions with nuclear overlap, i.e. non-single-diffractive collisions, is of particular interest. In this paper, we perform a calculation of exclusive J/$ψ$ photoproduction in non-single-diffractive p+p collisions at RHIC and LHC energies base on the pQCD motivated parametrization from world-wide experimental data, which could be further employed to improve the precision of phenomenal calculations for photoproduction in A+A collisions. The differential rapidity and transverse momentum distributions of J/$ψ$ from photoproduction are presented. In comparison with the J/$ψ$ production from hadronic interactions, we find that the contribution of photoproduction is negligible.

hep-ph↗